MiniMax H3 - Text to Video
ComfyUI in the Cloud
Deploy MiniMax H3 - Text to Video, from
$0.60 / hour
MiniMax-H3 Video Generation ComfyUI Workflow
A ComfyUI video generation workflow powered by MiniMax-H3, featuring an INT8-quantized diffusion model, Qwen3-VL 32B conditioning, dedicated video and audio autoencoders, and optional 8-step Turbo LoRA acceleration.
Workflow Overview
This workflow utilizes the MiniMax-H3 model for AI video generation within ComfyUI. It combines an INT8-quantized diffusion model, Qwen3-VL 32B vision-language conditioning, and separate video and audio autoencoders within a configurable generation subgraph.
The workflow includes configurable video dimensions, duration, and noise seed parameters. The displayed duration is set to 5.0 seconds, while width and height are exposed as external inputs controlled by connected workflow nodes.
An optional Turbo LoRA configuration is also included. The selected Turbo model is an 8-step variant with a configured model strength of 1.00. However, Turbo mode is disabled in the displayed configuration, so the selected acceleration settings are not currently enabled.
INT8 quantization helps reduce memory requirements for the primary diffusion model and video VAE. The workflow also includes an FP32 audio VAE and a Qwen3-VL 32B conditioning model identified with NVFP4 and AWQ quantization.
Generation Settings
Width
External Input
Height
External Input
Video Duration
5.0
Noise Seed
757358688076805
Turbo Mode
Disabled
Turbo Steps
8
Turbo Model Strength
1.00
Diffusion Precision
INT8
Model Configuration
Video Generation Model (UNet)
INT8minimax_h3_fl2va_pruned_int8_convrot
The MiniMax-H3 diffusion model serves as the primary video generation component. It is responsible for the denoising process used to generate video representations. The selected model is a pruned INT8-quantized configuration, helping reduce memory consumption compared with higher-precision model weights. The model receives conditioning information and generation parameters from other components within the complete workflow.
Text / Vision Conditioning Model
Qwen3-VL 32Bqwen3vl_32b_minimax_h3_nvfp4_awq
The Qwen3-VL 32B model provides vision-language conditioning for the MiniMax-H3 generation process. It is designed to interpret text instructions and visual information, producing representations that can guide the diffusion model. The selected model identifier includes NVFP4 and AWQ quantization designations. This conditioning component supports integration with compatible prompt-processing and visual conditioning nodes.
Video Variational Autoencoder (VAE)
INT8minimax_h3_video_vae_int8_convrot
The Video VAE provides the autoencoder component responsible for handling video latent representations. During video decoding, the autoencoder converts generated latent representations into visible video frames. The selected INT8-quantized configuration helps reduce memory requirements. This component can be connected to compatible video encoding and decoding nodes within the ComfyUI workflow.
Audio Variational Autoencoder (VAE)
FP32minimax_h3_audio_vae_fp32
The Audio VAE provides an additional autoencoder component for processing audio latent representations. Audio autoencoders support encoding and decoding operations between audio information and latent representations. The selected model uses FP32 precision, providing 32-bit floating-point processing. This component can be used with compatible audio generation, processing, and decoding nodes.
Optional Turbo LoRA
Disabledminimax_h3_fl2v_turbo_8step_v1.0_comfyui_bf16
The workflow includes an optional MiniMax-H3 Turbo LoRA designed for an 8-step generation configuration. The selected LoRA is identified as version 1.0 and uses BF16 precision. Its configured model strength is 1.00. Turbo mode is disabled in the displayed workflow, meaning this optional acceleration configuration is not currently enabled. When enabled through the appropriate workflow logic, the Turbo LoRA can modify the behavior of the primary diffusion model.
Turbo LoRA Configuration
The workflow includes an optional Turbo mode intended to support accelerated video generation. The selected Turbo LoRA is an 8-step variant with a configured model strength of 1.00. In the current configuration, Turbo mode is disabled.
| Turbo Mode | Disabled |
| Turbo LoRA | minimax_h3_fl2v_turbo_8step_v1.0_comfyui_bf16 |
| Turbo Model Strength | 1.00 |
| Turbo Steps | 8 |
| LoRA Precision | BF16 |
| LoRA Version | v1.0 |
The Turbo mode toggle determines whether the optional acceleration configuration is enabled. The strength parameter controls the configured LoRA influence when the Turbo model is applied. The Turbo steps parameter is set to 8, although this does not establish the actual sampling step count while Turbo mode is disabled.
Video Generation Configuration
The workflow exposes width, height, and duration parameters for configuring the generated video. Width and height are connected as external inputs, while duration is configured directly within the subgraph.
Video Width
External Input
Controls the horizontal dimensions of the generated video through a connected node.
Video Height
External Input
Controls the vertical dimensions of the generated video through a connected node.
Video Duration
5.0
Sets the configured duration used during video generation.
The duration parameter is set to 5.0, corresponding to a five-second configuration when interpreted in seconds by the internal nodes. The exact frame count and frame rate are not visible in the provided screenshot. Video dimensions depend on the connected width and height inputs.
Noise Seed Configuration
The workflow includes a configurable noise seed that controls random initialization during the video generation process. The displayed seed is 757358688076805.
| Noise Seed | 757358688076805 |
| Seed Control | Configurable |
| Randomization Control | Available |
A fixed seed can help reproduce video generation results when the model configuration, conditioning, and other generation parameters remain unchanged. Changing the seed can produce different video variations by modifying the random initialization used during generation.
Video and Audio Processing
The MiniMax-H3 workflow includes separate autoencoder components for video and audio. These components handle their respective latent representations during the generation and decoding process.
Video Autoencoder
minimax_h3_video_vae_int8_convrot
Precision: INT8
Processes video latent representations and supports decoding generated video data into visible frames. The INT8 configuration helps reduce memory consumption.
Audio Autoencoder
minimax_h3_audio_vae_fp32
Precision: FP32
Provides encoding and decoding capabilities for compatible audio processing components. The FP32 configuration uses 32-bit floating-point precision.
The inclusion of separate video and audio autoencoders provides support for processing both modalities within a larger workflow. The exact audio generation and synchronization behavior depends on the internal ComfyUI nodes and their connections.
Additional Parameters
| Diffusion Model | MiniMax-H3 FL2VA |
| Diffusion Precision | INT8 |
| Conditioning Model | Qwen3-VL 32B |
| Conditioning Quantization | NVFP4 / AWQ |
| Video VAE Precision | INT8 |
| Audio VAE Precision | FP32 |
| Duration | 5.0 |
| Noise Seed | 757358688076805 |
| Turbo Mode | Disabled |
| Turbo Steps | 8 (Configured) |
| Turbo Model Strength | 1.00 |
| Image Dimensions | External Width and Height Inputs |
Surrounding Nodes and Workflow Integration
The MiniMax-H3 subgraph consolidates model selection, generation duration, noise seed configuration, video dimensions, and optional Turbo acceleration settings into a reusable ComfyUI workflow component.
Width and height are exposed as external input connections, allowing upstream nodes to control the dimensions of generated video. The duration parameter determines the configured video length, while the noise seed controls random initialization.
The Qwen3-VL 32B conditioning model can process text instructions and visual information through compatible conditioning nodes. These conditioning representations can then guide the MiniMax-H3 diffusion model during the video generation process.
The optional Turbo configuration provides an additional LoRA model intended for accelerated generation. When Turbo mode is enabled, the workflow can apply the selected Turbo LoRA according to its internal node logic. In the displayed configuration, Turbo mode remains disabled.
The diffusion model performs the generation process using the supplied conditioning information and configured parameters. The video VAE can decode generated video latent representations into visible frames. The audio VAE provides a separate component for compatible audio decoding operations.
Generated video and audio data can be passed to downstream processing or export nodes within a complete ComfyUI workflow. Exact prompt connections, reference inputs, sampling settings, audio synchronization, and output nodes are not visible in the screenshot.
Workflow Highlights
- ✓ MiniMax-H3: Uses the MiniMax-H3 diffusion model for AI video generation.
- ✓ INT8 Quantization: Helps reduce diffusion model memory requirements compared with higher-precision weights.
- ✓ Qwen3-VL 32B: Integrates vision-language conditioning with NVFP4/AWQ model identifiers.
- ✓ Video Duration: Includes a configurable duration parameter currently set to 5.0.
- ✓ Optional Turbo Mode: Supports an 8-step Turbo LoRA configuration, currently disabled.
- ✓ Turbo LoRA Strength: Includes a configurable model strength set to 1.00.
- ✓ Video Autoencoder: Uses an INT8 video VAE for processing video latent representations.
- ✓ Audio Autoencoder: Includes an FP32 audio VAE for compatible audio processing operations.
- ✓ Flexible Video Dimensions: Width and height are controlled through external node connections.
- ✓ Seed-Based Generation: Supports configurable noise initialization with the displayed seed set to 757358688076805.
- ✓ Reusable Subgraph: Consolidates model selection, duration, Turbo settings, and generation parameters into a reusable workflow component.
Workflow Summary
This ComfyUI workflow combines the MiniMax-H3 INT8-quantized diffusion model with Qwen3-VL 32B conditioning, an INT8 video VAE, and an FP32 audio VAE. It includes configurable video dimensions, a duration value of 5.0, and a noise seed of 757358688076805.
The workflow also includes an optional MiniMax-H3 Turbo LoRA identified as an 8-step BF16 variant. The configured Turbo model strength is 1.00, and the Turbo step value is set to 8. Turbo mode is currently disabled.
Qwen3-VL 32B provides vision-language conditioning, while the MiniMax-H3 diffusion model performs the generation process. The separate video and audio autoencoders support processing their respective latent representations.
The subgraph organizes model configuration, generation duration, noise initialization, optional Turbo acceleration, and video dimensions for integration into larger ComfyUI workflows.